Fig. 1

Illustration of the Lisa framework. a The peak-RP score models the effect of TR binding sites on the regulation of a gene. TR binding sites are binary values, and peaks nearer to the gene’s TSS have a greater influence than ones further away. b The chrom-RP score summarizes the effect of the DNase-seq or H3K27ac chromatin environment on a gene. The chrom-RP score is based on a continuous rather than a binary signal quantification. c Overview of the Lisa framework. (1) H3K27ac ChIP-seq or DNase-seq data from the Cistrome DB is summarized using the chrom-RP score for each gene. (2) H3K27ac ChIP-seq or DNase-seq samples that can discriminate between the query gene set and the background gene set are selected, and the regression parameters define a chrom-RP model. (3) Each TR cistrome from the Cistrome DB is evaluated as a putative regulator of the query gene set through in silico deletion, which involves the elimination of H3K27ac ChIP-seq or DNase-seq signal at the binding sites of the putative regulator. (4) The chrom-RP model, based on in silico deletion signal, is compared to the model without deletion for each gene in the query and background gene sets. A p value is calculated using the Wilcoxon rank test comparison of the query and background ΔRPs. (5) The peak-RP based on TR ChIP-seq peaks is calculated for the putative regulatory cistrome, and the statistical significance of peak-RP distributions from the query and background gene sets is calculated. (6) p values from the H3K27ac ChIP-seq, DNase-seq, and peak-RP analysis are combined using the Cauchy combination test. TR cistromes are ranked based on the combined p value